Keras/TensorFlow GPU -- 停留在 Epoch 1

Question

我正在努力设置一个简单的深度学习代码以在我的 GPU 上运行。

该代码是来自 cifar10 数据集的简单 CNN（从https://machinelearningmastery.com/how-to-develop-a-cnn-from-scratch-for-cifar-10-photo-classification/复制/过去）。

当我检查它是否找到 GPU 时：

from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())

我得到：

[name: "/device:CPU:0"
device_type: "CPU"
memory_limit: 268435456
locality {
}
incarnation: 1854574019269825039
, name: "/device:GPU:0"
device_type: "GPU"
memory_limit: 23174440576
locality {
  bus_id: 1
  links {
  }
}
incarnation: 269960360402377625
physical_device_desc: "device: 0, name: GeForce RTX 3090, pci bus id: 0000:65:00.0, compute capability: 8.6"
]

它确实找到了 cuda 和 cudnn 但卡在 Epoch 1：

2021-04-11 22:09:18.343347: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:116] None of the MLIR optimization passes are enabled (registered 2)
2021-04-11 22:09:18.363546: I tensorflow/core/platform/profile_utils/cpu_utils.cc:112] CPU Frequency: 3699850000 Hz
Epoch 1/100
2021-04-11 22:09:18.915658: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.11
2021-04-11 22:09:19.399116: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublasLt.so.11
2021-04-11 22:09:19.401652: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudnn.so.8

如果你有任何想法？ 提前致谢

Answer 1

不要使用 pip 版本，而是使用 conda 我这样做并为我工作：

conda create -n tf-gpu
conda activate tf-gpu
conda install -c conda-forge tensorflow-gpu

然后安装nessecary包